Life evaluation model establishment method and device, equipment and storage medium

Abstract

The invention relates to a life evaluation model establishment method, device and equipment and a storage medium, and is applied to the field of integrated circuits, and the method comprises the steps: determining the electric stress condition of a power device, and measuring the failure time of the power device under the electric stress condition; calculating a Webster cumulative distribution function of the power device in the failure time; fitting and calculating a characteristic life value of the power device under the electric stress condition according to the Webster cumulative distribution function and a preset global shape factor; and constructing a deep neural network model, training the deep neural network model by using the characteristic life value obtained by fitting calculation, and generating an accelerated life evaluation model. The method has the technical effects that failure analysis and service life evaluation are carried out on devices with complex processes and structures, and the evaluation accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of integrated circuit technology, and in particular to a method, device, equipment and storage medium for establishing a life assessment model. Background Art

[0002] With the continuous progress of semiconductor technology, solid-state power electronics technology has been developing rapidly. The development of solid-state devices not only promotes the technological revolution of the information industry, but also encourages people to think and live in new ways. As the basic discipline of semiconductor technology, semiconductor materials play a vital role in the development of energy and information technology.

[0003] In the process of semiconductor technology development, GaN (gallium nitride) devices are the best choice for high-frequency and high-efficiency power modules because of their higher electron saturation drift velocity and electron mobility. Natural GaN is a depletion-mode device, and the realization of enhanced GaN usually involves the realization of a metal-insulator-semiconductor (MIS) structure and a P-type gate structure, which may bring additional reliability issues to the device.

[0004] When realizing the metal-insulator-semiconductor (MIS) structure, the ultraviolet rays in the plasma cause serious damage to the semiconductor surface due to the long plasma irradiation time. The surface damage will lead to increased leakage current, poor threshold voltage (Vth) stability and current collapse (increased dynamic on-resistance). However, p-type gate gallium nitride (p-GaN) has forward gate overvoltage failure and time-related breakdown. When a high electrical stress voltage is applied to the gate, a large voltage drop and electric field will appear in the p-GaN depletion region near the metal interface, thereby promoting the formation of a penetration path. This degradation mechanism is consistent with time-related dielectric breakdown.

[0005] Time-dependent dielectric breakdown (TDDB) has been extensively studied in power devices of silicon and other processes, and TDDB has become a necessary evaluation method for studying transistor reliability. However, during the dielectric breakdown process, dielectric layer breakdown may occur, which will lead to device destruction and loss of switching function, resulting in power module failure. Therefore, accurately evaluating the life of power devices and analyzing the failure mechanism of new structures have become the key to improving device reliability and module quality. At present, the commonly used accelerated life assessment models for power devices are exponential function models and power function models. By fitting the characteristic life with the exponential function model and the power function model, the service life under the rated operating voltage or the typical operating voltage under the rated service life can be obtained.

[0006] However, traditional exponential function models and power function models can only fit the characteristic life under a fixed distribution mode. With the improvement of system indicators and the increase of breakdown voltage, the failure mechanism and life characteristics of new device processes and structures developed have become complex, and it is difficult to accurately evaluate the failure mechanism and life through traditional models. Summary of the invention

[0007] In order to help solve the problem that traditional life assessment models are difficult to accurately assess the failure mechanism and life of power devices, the present application provides a life assessment model establishment method, device, equipment and storage medium.

[0008] In a first aspect, the present application provides a method for establishing a life assessment model, which adopts the following technical solution: the method is applied to a life assessment model establishment system, the life assessment model establishment system includes a power device, and the method includes:

[0009] determining an electrical stress condition of the power device according to a gate leakage current of the power device, and measuring a failure time of the power device under the electrical stress condition;

[0010] Calculate the Wei's cumulative distribution function of the power device at the failure time;

[0011] Calculate the characteristic life value of the power device under the electrical stress condition according to the Weiss cumulative distribution function and the preset global shape factor fitting;

[0012] A deep neural network model is constructed, and the characteristic life value obtained by fitting calculation is used to train the deep neural network model to generate an accelerated life assessment model, where the accelerated life assessment model is a partitioned deep neural network model.

[0013] In a specific implementation manner, determining the electrical stress condition of the power device according to the gate leakage current of the power device and measuring the failure time of the power device under the electrical stress condition comprises:

[0014] Measuring a steep increase point of the gate leakage current of the power device, and determining a steep increase stress range of the gate leakage current according to the steep increase point;

[0015] Determining a number of different electrical stress conditions according to the steep stress range;

[0016] The dielectric breakdown full life acceleration experiment is used to measure the failure time of the power device under several electrical stress conditions.

[0017] In a specific implementation scheme, the calculation method of the Wei's cumulative distribution function includes:

[0018]

[0019] Wherein, t represents the failure time, β represents the shape factor, η represents the expansion factor, γ represents the delay, and F(t) represents the Weiss cumulative distribution function.

[0020] In a specific implementation scheme, the calculation of the characteristic life value of the power device under the electrical stress condition according to the Weiss cumulative distribution function and the preset global shape factor fitting includes:

[0021] Fitting the Wei's distribution of the failure time according to the Wei's cumulative distribution function;

[0022] Calculating the intercept value of the failure time under the electrical stress condition according to the Wei's distribution of the failure time;

[0023] Calculating a characteristic life value of the power device under the electrical stress condition according to the preset global shape factor and the intercept value;

[0024] The calculation method of the characteristic life value includes:

[0025]

[0026] Wherein, intercept_j represents the calculated intercept value, β′ represents the global shape factor, and η_j represents the characteristic life value under the jth set of stress conditions.

[0027] In a specific implementation scheme, the deep neural network model is constructed, and the characteristic life value obtained by fitting calculation is used to train the deep neural network model to generate an accelerated life assessment model, which includes:

[0028] Constructing the input layer, hidden layer and output layer of the deep neural network model, and setting activation functions, weights and biases in the hidden layer;

[0029] Determining a demarcation point of the electrical stress condition according to a steep increase point of the gate leakage current, and generating a low electrical stress condition and a high electrical stress condition;

[0030] Determine the range of the electrical stress condition corresponding to the characteristic life value, and convert the format of the characteristic life value according to different electrical stress conditions;

[0031] If the electrical stress condition corresponding to the characteristic life value is a low electrical stress condition, converting the characteristic life value into a real number format and generating a real number characteristic life value;

[0032] If the electrical stress condition corresponding to the characteristic life value is a high electrical stress condition, converting the characteristic life value into a dB unit format and generating a logarithmic characteristic life value;

[0033] Normalizing the real characteristic life value and the logarithmic characteristic life value to generate a real normalized characteristic life value and a logarithmic normalized characteristic life value;

[0034] Determining a range of electrical stress conditions corresponding to the characteristic life value, and training the deep neural network model according to different electrical stress conditions;

[0035] If the electrical stress condition corresponding to the characteristic life value is a low electrical stress condition, the deep neural network model is trained using the real normalized characteristic life value to generate a real neural network model;

[0036] If the electrical stress condition corresponding to the characteristic life value is a high electrical stress condition, the deep neural network model is trained using the logarithmic normalized characteristic life value to generate a logarithmic neural network model;

[0037] The real number neural network model and the logarithmic neural network model are set as the accelerated lifetime assessment model.

[0038] In a specific implementation scheme, the deep neural network model is constructed, and the characteristic life value obtained by fitting calculation is used to train the deep neural network model to generate an accelerated life assessment model, which includes:

[0039] Constructing the input layer, hidden layer and output layer of the deep neural network model, and setting activation functions, weights and biases in the hidden layer;

[0040] Converting the characteristic life value into a real number format and a dB unit format respectively, and generating a real number characteristic life value and a logarithmic characteristic life value after the format conversion;

[0041] Normalizing the real number characteristic life value and the logarithmic characteristic life value after format conversion, and generating a real number normalized characteristic life value and a logarithmic normalized characteristic life value;

[0042] Using the real normalized characteristic life value to train the deep neural network model and generate a real neural network model;

[0043] Using the logarithmic normalized characteristic life value to train the deep neural network model and generate a logarithmic neural network model;

[0044] According to the training results of the real neural network model and the logarithmic neural network model, the electrical stress condition is divided into a low electrical stress condition and a high electrical stress condition;

[0045] Under the low electrical stress condition, the real number neural network model is used accordingly, and under the high electrical stress condition, the logarithmic neural network model is used accordingly, and the final accelerated life assessment model is generated.

[0046] In a specific implementation scheme, after generating the accelerated life assessment model, the method further includes:

[0047] The results output by the accelerated life assessment model are subjected to denormalization processing.

[0048] In a second aspect, the present application provides a device for establishing a life assessment model, which adopts the following technical solution: the device is applied to a life assessment model establishment system, the life assessment model establishment system includes a power device, and the device includes:

[0049] a failure time calculation module, used to determine the electrical stress condition of the power device according to the gate leakage current of the power device, and measure the failure time of the power device under the electrical stress condition;

[0050] A distribution function calculation module, used to calculate the Weiss cumulative distribution function of the power device at the failure time;

[0051] A characteristic life calculation module, used for calculating the characteristic life value of the power device under the electrical stress condition according to the Weiss cumulative distribution function and a preset global shape factor fitting;

[0052] The model construction training module is used to construct a deep neural network model, and use the characteristic life value obtained by fitting calculation to train the deep neural network model to generate an accelerated life assessment model, which is a partitioned deep neural network model.

[0053] In a third aspect, the present application provides a computer device, which adopts the following technical solution: it includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and execute any of the life assessment model establishment methods described above.

[0054] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: storing a computer program that can be loaded by a processor and execute any of the above-mentioned life assessment model establishment methods.

[0055] In summary, this application has the following beneficial technical effects:

[0056] By first measuring the basic parameters of the power device, the deep neural network model is trained using the measured parameters to obtain a model that can evaluate the life of the power device. For power devices with complex process structures, the model obtained by this modeling can also be used to accurately evaluate the life and reliability of the device, so as to solve the shortcomings of the existing GaN power device life model. Secondly, through the method of constructing a deep neural network model, the accuracy of fitting the life curve of GaN power devices can be improved, the reliability characteristics before and after the steep increase point of the gate leakage current can be predicted, and the formation of the gate leakage current channel can be judged, providing a basis for reliability prediction and design. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a flow chart of a method for establishing a life assessment model in an embodiment of the present application;

[0058] Figure 2 This is a schematic diagram used in the embodiment of the present application to illustrate the situation where the voltage of the device increases sharply within the range of 7V-8V;

[0059] Figure 3 This is a schematic diagram of the results of the dielectric breakdown full accelerated life test in the embodiment of the present application;

[0060] Figure 4 is a schematic diagram used in the embodiment of the present application to illustrate the results of fitting the life of a power device with a Weiss distribution under different electrical stress conditions;

[0061] Figure 5 is a schematic diagram of the deep neural network model structure in an embodiment of the present application;

[0062] Figure 6 It is a schematic diagram of the results of simulating power and device life of the traditional exponential function model and the power function model in the embodiment of the present application;

[0063] Figure 7 This is a schematic diagram of the result of the simulated power and device life of the evaluation model constructed by this solution in the embodiment of the present application;

[0064] Figure 8 is a schematic diagram of a device for establishing a life assessment model in an embodiment of the present application;

[0065] Fig. 9 It is a schematic diagram used to embody a computer device in an embodiment of the present application.

[0066] Attached figure numbers: 801, failure time calculation module; 802, distribution function calculation module; 803, characteristic life calculation module; 804, model building training module. DETAILED DESCRIPTION

[0067] The following combination Figure 1-Figure 9 This application is described in further detail.

[0068] The embodiment of the present application discloses a method for establishing a life assessment model. The model establishment method can be used to construct a life assessment model with a wide range of applications, so that the model can be used to perform failure analysis and life prediction of power devices. With the continuous advancement of semiconductor technology, solid-state power electronics technology has continued to develop rapidly. While the development of solid-state devices is promoting the development of technological revolution in the information industry, it is also prompting people to think and live in new ways. As a basic discipline of semiconductor technology, semiconductor materials play a vital role in the development of energy and information technology.

[0069] The evolution of semiconductor technology can be summarized into three important stages, namely the first generation of semiconductor materials represented by silicon (Si), the second generation of semiconductor materials represented by gallium arsenide (GaAs), and the third generation of semiconductor materials represented by gallium nitride (GaN) and silicon carbide (SiC). Compared with other materials, GaN is more suitable for high-frequency and high-power scenarios, especially in the MHz frequency band. Because GaN has a higher electron saturation drift velocity and electron mobility, it is the best choice for high-frequency and high-efficiency power modules.

[0070] Epitaxial GaN power devices on silicon (Si) substrates have lattice mismatch and thermal mismatch, which will produce fatal defects in the GaN layer, so complex buffer layers and epitaxial layers are required; in addition, natural GaN is a depletion-mode device, and the realization of enhancement-mode GaN usually involves the realization of a metal-insulator-semiconductor (MIS) structure and a P-type gate structure, which will also bring additional reliability issues to the device.

[0071] When realizing the metal-insulator-semiconductor (MIS) structure, due to the long plasma irradiation time, the ultraviolet light from the plasma causes serious damage to the semiconductor surface. The surface damage will lead to increased leakage current, poor threshold voltage (Vth) stability and current collapse (increased dynamic on-resistance). However, p-type gate gallium nitride (p-GaN) has forward gate overvoltage failure and time-related breakdown. When a high electrical stress voltage is applied to the gate, a large voltage drop and electric field will appear in the p-GaN depletion region near the metal interface, thereby promoting the formation of a penetration path. This degradation mechanism is consistent with time-related dielectric breakdown.

[0072] Time-dependent dielectric breakdown (TDDB) has been extensively studied in power devices of silicon and other processes, and TDDB has become a necessary evaluation method for studying transistor reliability. However, during the dielectric breakdown process, dielectric layer breakdown may occur, which will lead to device destruction and loss of switching function, resulting in power module failure. Therefore, accurately evaluating the life of power devices and analyzing the failure mechanism of new structures have become the key to improving device reliability and module quality. At present, the commonly used accelerated life evaluation models for power devices are exponential function models and power function models. By fitting the characteristic life with the exponential function model and the power function model, the service life under the rated working voltage or the typical working voltage under the rated service life can be obtained. For traditional devices, three characteristic lifespans are generally obtained through full life accelerated tests, and these three characteristic lifespans are basically distributed on the same straight line under the log coordinates. Therefore, the exponential function and power function models can obtain a good fitting effect and the minimum error function.

[0073] However, the traditional exponential function model and power function model can only fit the characteristic life under the fixed distribution mode. With the improvement of system indicators and the increase of breakdown voltage, the failure mechanism and life characteristics of the new device process and structure developed become complicated, and it is difficult to accurately evaluate the failure mechanism and life through the traditional model; in addition, the traditional model has low accuracy in fitting the accelerated life under high electrical stress conditions, and the fitting effect is not good. In order to solve the problems existing in the traditional model in calculating the life of power devices, the present application provides a method for establishing a life assessment model.

[0074] Reference Figure 1 , the method comprises the following steps:

[0075] S10, determining an electrical stress condition of the power device according to a gate leakage current of the power device, and measuring a failure time of the power device under the electrical stress condition.

[0076] Specifically, when establishing a life assessment model, it is necessary to first obtain the relevant parameters of the power device in order to facilitate the learning and training of the model. Performing full life acceleration experiments on power devices can obtain the parameters of the failure time of the power device, that is, the time it takes for the power device to completely fail under certain stress conditions, and then use this parameter to build and train the model. Under different electrical stress conditions, the failure time of the device is different.

[0077] The specific method of measuring the failure time of the power device may include: first, measuring the steep increase point of the gate leakage current of the power device, and determining the steep increase stress range of the gate leakage current according to the steep increase point; specifically, the Kelvin test method may be used to measure the gate DC voltage of the device in a step-by-step manner within a preset voltage range through a semiconductor parameter analyzer, for example, the source DC voltage Vs of the device is 0V, the drain DC voltage Vd is 0.1V, and the gate DC voltage Vg is swept from 0V to 15V in steps of 0.05V, referring to Figure 2 , the leakage current increases sharply when the gate voltage stress is in the range of 7V-8V, indicating that there are different gate dielectric degradation mechanisms at this point, so the range can be set to [7V, 8.3V].

[0078] After that, several different electrical stress conditions are determined according to the steep stress range. The set stress conditions are divided into several groups. The specific number of condition groups can be set by the user, and there is no limit here. For example, in the range of [7V, 8.3V], 9 groups of stress conditions are selected, and more than 8 identical devices are tested under each group of stress conditions. Finally, the dielectric breakdown full life accelerated experiment is used to measure the failure time of the power device under several electrical stress conditions. For example, refer to Figure 3 , Vg=7.1V, Vd=0.1V of the power device, the gate leakage current changes with time, the current increases slowly with time, and noise appears at the same time. When the time is greater than 324s, the current increases sharply, indicating the formation of a penetration channel in the dielectric, and the hard breakdown causes the device to fail. The breakdown time at this time is recorded, and the same method is used to calculate the breakdown time of each power device under the corresponding stress conditions, and the sorting and numbering are performed. The sorting method is not limited. In the present application scheme, the calculated failure times are sorted and numbered from small to large.

[0079] S20, calculating the Weiss cumulative distribution function of the power device at failure time.

[0080] Specifically, multiple groups of electrical stress conditions were set during the experiment to observe the failure time or breakdown time of power devices under different electrical stress conditions. Weibull distribution, also known as Weibull distribution, is a continuous probability distribution that is widely used in reliability analysis, life testing and other fields. By calculating the Weibull cumulative distribution function of the power device under different electrical stress conditions, it is convenient to calculate the characteristic life value of the power device under different electrical stress conditions. The failure times obtained from the original measurement are arranged in random order. The measured failure times are first sorted by size. In the embodiment of the present application, they are sorted from small to large in terms of failure time. Each power device has a corresponding measured failure time and number. When calculating the device life, the shape factor and characteristic life value can be extracted by calculating the distribution cumulative function. The Weibull cumulative distribution function can be expressed as:

[0081]

[0082] Among them, t represents the failure time, β represents the shape factor, η represents the expansion factor, which can also be understood as the failure rate, γ represents the delay, and F(t) represents the Weiss cumulative distribution function.

[0083] Among them, there are many ways to calculate the value F(t) of the Wei's cumulative distribution function, and the median rank method has a relatively better calculation effect. In the embodiment of the present application, the calculation of the median rank method is taken as an example for explanation. The calculation method of the median rank method can be expressed as:

[0084]

[0085] Among them, i represents the device number after sorting under a certain set of electrical stress conditions, that is, the number of the power device corresponding to the failure time t in the above distribution cumulative function; F(i) represents the Wei's distribution cumulative function value of the i-th power device under a certain set of electrical stress conditions, which can also be understood as the probability of the penetration path being formed at a certain time under given stress conditions; n represents the total number of power devices under each set of electrical stress conditions. The calculated value of F(i) is the value of F(t).

[0086] S30, calculating a characteristic life value of the power device under electrical stress conditions according to the Weiss cumulative distribution function and a preset global shape factor fitting.

[0087] Specifically, the characteristic life value of the power device under different electrical stress conditions can be obtained by calculating and fitting according to the Weiss cumulative distribution function and the preset global shape factor. The specific calculation method may include: first, fitting the Weiss distribution of the failure time according to the Weiss cumulative distribution function; when the delay parameter γ in the Weiss cumulative distribution function is 0, the Weiss cumulative distribution function can be expressed as:

[0088] ln[-ln(1-F(t))]=βln(t)-βln(η),

[0089] Where t is the failure time, β is the shape factor, η is the expansion factor, and F(t) is the Weiss cumulative distribution function.

[0090] By fitting ln[-ln(1-F(t))] and ln(t), a straight line with slope β and intercept -βln(η) can be generated. In the semi-logarithmic diagram, the fitting result can be called Wei's distribution.

[0091] Afterwards, the intercept value of the failure time under electrical stress conditions is calculated based on the Weiss distribution of the failure time. Specifically, the fitting calculation formula of the Weiss distribution can be expressed as:

[0092] y_j=β·x_j[1:n]+intercept_j,

[0093] Among them, x_j[1:n] represents the natural logarithm of the failure time from the 1st to the nth device under the jth set of stress conditions, intercept_j represents the intercept under each set of stress conditions, and y_j represents the value of ln[-ln(1-F(t))]. Assuming that the initial value of β is set to 1 and the initial value of intercept_j is set to 0, the global slope β and the intercept value under each stress can be obtained by fitting through the least squares method. The results obtained by fitting the Weiss distribution refer to Figure 4 As shown, the global shape factor β set in the present application is 1.25. In practical applications, the global shape factor can be adjusted and determined according to the actual fitting results of the user, and is not limited here.

[0094] Finally, the characteristic life value of the power device under electrical stress conditions is calculated according to the preset global shape factor and intercept value;

[0095] The characteristic life value is calculated by:

[0096]

[0097] Where intercept_j represents the calculated intercept value, β′ represents the global shape factor, and η_j represents the characteristic life value under the jth set of stress conditions. intercept_j corresponds to -βln(η) in the Weiss cumulative distribution function, and the characteristic life η_j corresponds to the calculated value when ln[-ln(1-F(t))]=0.

[0098] S40, constructing a deep neural network model, and using the characteristic life value obtained by fitting calculation to train the deep neural network model, to generate an accelerated life assessment model, where the accelerated life assessment model is a partitioned deep neural network model.

[0099] Specifically, a basic deep neural network model is constructed, and then the initial deep neural network model is learned and trained according to the characteristic life values ​​calculated under different electrical stress conditions of the power device to generate an accelerated life assessment model that can eventually be used for evaluation; the model training is performed separately under different electrical stress conditions, so that the final accelerated life assessment model is a partitioned deep neural network model.

[0100] In the present application, the basic parameters of the power device are first measured, and the deep neural network model is trained using the measured parameters to obtain a model that can evaluate the life of the power device. For power devices with complex process structures, the model obtained by this modeling can also be used to accurately evaluate the life and reliability of the device, so as to solve the shortcomings of the existing GaN power device life model. Secondly, through the method of constructing a deep neural network model, the accuracy of fitting the life curve of GaN power devices can be improved, the reliability characteristics before and after the steep increase point of the gate leakage current can be predicted, and the formation of the gate leakage current channel can be judged, providing a basis for reliability prediction and design.

[0101] In one embodiment, a deep neural network model is constructed, and the characteristic life value obtained by fitting calculation is used to train the deep neural network model. The method of generating an accelerated life assessment model can be specifically performed as follows:

[0102] First, construct the input layer, hidden layer, and output layer of the deep neural network model, and set the activation function in the hidden layer. A deep neural network is a complex network structure composed of multiple layers, which usually include input layer, hidden layer, and output layer. The hidden layer is a key component in the deep neural network. The hidden layer is located between the input layer and the output layer. A deep neural network can contain multiple hidden layers, which are stacked in sequence, and the output of each layer is used as the input of the next layer; refer to Figure 5In the embodiment of the present application, the constructed deep neural network model includes an input layer, an output layer and three hidden layers, each hidden layer includes 16 neurons, and the main function of the hidden layer is to extract the features of the input data; the number of hidden layers of the constructed deep neural network and the number of neurons contained in each hidden layer can be set by the user according to the actual modeling requirements, and the number of neurons set can be set to suit the model operation. If the number of neurons is too small, the accuracy of the calculation fitting will be low and the fitting effect will be poor. If the number of neurons is more, the fitting accuracy will be higher, but the network will take longer to calculate the complexity, and overfitting may occur. Therefore, the number of neurons should not be too many or too few. 16 neurons are used in the present application; the three hidden layers and each hidden layer containing 16 neurons in the embodiment of the present application are only used as examples for explanation, and are not limited here. Activation functions, weights and biases are set in each hidden layer. Through different weights and activation functions, nonlinear transformation is performed on the input data to capture the abstract features of different levels of the input data. The user can select and adjust the activation function according to the actual computing requirements and the actual modeling effect. In the embodiment of the present application, the activation function is explained by taking the Sigmoid function as an example. The calculation method of the Sigmoid function can be expressed as:

[0103]

[0104] Among them, z represents the weighting function, and the weighting function z can be expressed as:

[0105] z=Σw i x i +b,

[0106] Among them, x i represents the independent variable of the neural network. In the embodiment of the present application, x of the first hidden layer i That is, the electrical stress condition of the input layer, and x of the second hidden layer i That is, z is the output of the first hidden layer, and x is the output of the third hidden layer. i That is z, the output of the second hidden layer; w i represents weight and b represents bias.

[0107] After building the basic neural network model, the input training data is processed to distinguish the electrical stress conditions into high and low electrical stress conditions. The specific distinction can be determined by determining the demarcation point of the electrical stress conditions based on the steep increase point of the gate leakage current to obtain low electrical stress conditions and high electrical stress conditions. Figure 6When fitting the accelerated life under high electrical stress conditions, the traditional exponential function model and power function model have lower accuracy in high stress fitting, smaller dynamic range, lower fitting accuracy and poor fitting effect because more error weights are allocated under low stress conditions. Therefore, by distinguishing between high and low electrical stress conditions, the fitting accuracy of the overall accelerated life of the power device can be improved, thereby improving the accuracy of determining the failure mechanism of the power device.

[0108] Afterwards, the range of the electrical stress condition corresponding to the characteristic life value is determined, and the format of the characteristic life value is converted according to different electrical stress conditions; if the electrical stress condition corresponding to the characteristic life value is a low electrical stress condition, the characteristic life value is converted into a real number format, and a real characteristic life value is generated. The conversion method can be expressed as:

[0109] η_j_s=η_j,

[0110] Among them, η_j represents the characteristic life value calculated by Wei's distribution, and η_j_s represents the real characteristic life value after real number conversion.

[0111] If the electrical stress condition corresponding to the characteristic life value is a high electrical stress condition, the characteristic life value is converted into a dB unit format and a logarithmic characteristic life value is generated. The conversion method can be expressed as:

[0112] η_j_dB=20log(abs(η_j)),

[0113] Among them, η_j represents the characteristic life value calculated by Wei's distribution, and η_j_dB represents the logarithmic characteristic life value after conversion to dB unit format.

[0114] Referring to Table 1, there is a difference of three orders of magnitude between the characteristic life under high stress and low stress conditions. This will cause the high-order characteristic life to contribute more to the weight of the error function during training, resulting in a decrease in the fitting accuracy of the low-order characteristic life. Therefore, under low stress conditions, training is carried out in real number format, and under high stress conditions, training is carried out in an algorithm format with dB as the unit.

[0115] Table 1, characteristic life under different stress conditions

[0116]

[0117] After that, the characteristic life value after format conversion is normalized, that is, the real characteristic life value and the logarithmic characteristic life value are normalized, and the real normalized characteristic life value and the logarithmic normalized characteristic life value are generated. The normalization method can be expressed as:

[0118]

[0119] Among them, η_jx represents the real characteristic life value and logarithmic characteristic life value before normalization, η_x min Indicates the minimum characteristic life, η_x max Indicates the maximum characteristic life, η_jx norm Represents the normalized real-number normalized characteristic life value and the logarithmic normalized characteristic life value.

[0120] After normalization, the range of the electrical stress condition corresponding to the characteristic life value is determined, and the deep neural network model is trained according to different electrical stress conditions; if the electrical stress condition corresponding to the characteristic life value is a low electrical stress condition, the real normalized characteristic life value is used to train the deep neural network model, and a real neural network model is generated; if the electrical stress condition corresponding to the characteristic life value is a high electrical stress condition, the logarithmic normalized characteristic life value is used to train the deep neural network model, and a logarithmic neural network model is generated. Finally, the real neural network model and the logarithmic neural network model are set as accelerated life assessment models, and the accelerated life assessment model finally constructed can be expressed as:

[0121]

[0122] Among them, Vkink represents the dividing point between high and low electrical stress conditions, Dnn_dB represents the logarithmic neural network model trained under high electrical stress conditions, Dnn_s represents the real neural network model trained under low electrical stress conditions, and Vgs_j represents the actual applied electrical stress conditions.

[0123] When building the model, normalized data is used for learning and training. Considering that the normalized data cannot truly reflect the actual life value, after the model is built, in practical applications, the data actually output by the model needs to be denormalized and the denormalized data needs to be converted to a real number format to obtain the true characteristic life value of the power device. The conversion to the real number format means that the real characteristic life value does not require additional processing, and the logarithmic characteristic life value, that is, the value in dB unit format is converted to a real value.

[0124] In one embodiment, a deep neural network model is constructed, and the characteristic life value obtained by fitting calculation is used to train the deep neural network model. The method of generating an accelerated life assessment model can be specifically performed as follows:

[0125] First, construct the input layer, hidden layer, and output layer of the deep neural network model, and set the activation function in the hidden layer. A deep neural network is a complex network structure composed of multiple layers, which usually include input layer, hidden layer, and output layer. The hidden layer is a key component in the deep neural network. The hidden layer is located between the input layer and the output layer. A deep neural network can contain multiple hidden layers, which are stacked in sequence, and the output of each layer is used as the input of the next layer; refer to Figure 5 In the embodiment of the present application, the constructed deep neural network model includes an input layer, an output layer and three hidden layers, each hidden layer includes 16 neurons, and the main function of the hidden layer is to extract the features of the input data; the number of hidden layers of the constructed deep neural network and the number of neurons contained in each hidden layer can be set by the user according to the actual modeling needs. Too few or too many neurons will affect the fitting accuracy of the network. Therefore, this application uses 16 neurons. The three hidden layers and each hidden layer containing 16 neurons in the embodiment of the present application are only used as examples for explanation, and are not limited here. An activation function is set in each hidden layer. Through different weights and activation functions, the input data is nonlinearly transformed, so as to capture the abstract features of the input data at different levels. The selection of the activation function can be selected and adjusted by the user according to the actual computing needs and the actual modeling effect. In the embodiment of the present application, the activation function is explained by taking the Sigmoid function as an example. The calculation method of the Sigmoid function can be expressed as:

[0126]

[0127] Among them, z represents the weighting function, and the weighting function z can be expressed as:

[0128] Z=Σw i x i +b,

[0129] Among them, x i represents the independent variable of the neural network. In the embodiment of the present application, x of the first hidden layer i That is, the electrical stress condition of the input layer, and x of the second hidden layer i That is, z is the output of the first hidden layer, and x is the output of the third hidden layer. i That is z, the output of the second hidden layer; w i represents weight and b represents bias.

[0130] Reference Figure 6Considering that the traditional exponential function model and power function model have low fitting accuracy when fitting the accelerated life under high electrical stress conditions, and because the error weight is mostly allocated to low stress conditions, the high stress fitting accuracy is low, the dynamic range is small, and the fitting effect is poor, therefore, by distinguishing high and low electrical stress conditions, the fitting accuracy of the overall accelerated life of the power device can be improved, thereby improving the accuracy of determining the failure mechanism of the power device. In addition to determining the steep increase point of the gate leakage current, the division of high and low electrical stress conditions can also be determined by first performing global model training, and then determining the dividing point based on the results of the model training.

[0131] Specifically, referring to Table 1 above, there is a difference of three orders of magnitude in the characteristic life under high stress and low stress conditions. This will cause the high-order characteristic life to contribute more to the weight of the error function during training, resulting in a decrease in the fitting accuracy of the low-order characteristic life. At the same time, the real number format and the algorithm format with dB as the unit are used for training. Finally, the high and low dividing points can be distinguished according to the training results.

[0132] First, the characteristic life value is converted into real number format and dB unit format respectively, and the real number characteristic life value and logarithmic characteristic life value after format conversion are generated; the real number format conversion method can be expressed as:

[0133] η_j_s=η_j,

[0134] Among them, η_j represents the characteristic life value calculated by Weiss distribution, and η_j_s represents the real characteristic life value after real number conversion. At this time, all electrical stress conditions are converted into real number formats, regardless of high or low;

[0135] The conversion to logarithmic format can be expressed as:

[0136] η_j_dB=20log(abs(η_j)),

[0137] Among them, η_j represents the characteristic life value calculated by Weiss distribution, and η_j_dB represents the logarithmic characteristic life value after the format conversion of dB unit. Similarly, the logarithmic format conversion is performed on all electrical stress conditions, regardless of high or low.

[0138] Afterwards, the real characteristic life value and the logarithmic characteristic life value after the format conversion are normalized to generate a real normalized characteristic life value and a logarithmic normalized characteristic life value. The normalization method can be expressed as:

[0139]

[0140] Among them, η_jx represents the real characteristic life value and logarithmic characteristic life value before normalization, η_x minIndicates the minimum characteristic life, η_x max Indicates the maximum characteristic life, η_jx norm Represents the normalized real-number normalized characteristic life value and the logarithmic normalized characteristic life value.

[0141] After the data processing is completed, the real normalized characteristic life value is used to train the deep neural network model and generate a real neural network model; at the same time, the logarithmic normalized characteristic life value is used to train the deep neural network model and generate a logarithmic neural network model.

[0142] Afterwards, according to the training results of the real neural network model and the logarithmic neural network model, the electrical stress conditions are divided into low electrical stress conditions and high electrical stress conditions; after the two training methods are completed, there will be an overlapping area between the two models, and the distinction point between high and low electrical stress conditions can be set in the overlapping area. Finally, the real neural network model is used under low electrical stress conditions, and the logarithmic neural network model is used under high electrical stress conditions, and the final accelerated life assessment model is generated. The final constructed accelerated life assessment model can be expressed as:

[0143]

[0144] Among them, Vkink represents the dividing point between high and low electrical stress conditions, Dnn_dB represents the logarithmic neural network model trained under high electrical stress conditions, Dnn_s represents the real neural network model trained under low electrical stress conditions, and Vgs_j represents the actual applied electrical stress conditions.

[0145] When building the model, normalized data is used for learning and training. Considering that the normalized data cannot truly reflect the actual life value, after the model is built, in practical applications, the data actually output by the model needs to be denormalized and the denormalized data needs to be converted to a real number format to obtain the true characteristic life value of the power device. The conversion to the real number format means that the real characteristic life value does not require additional processing, and the logarithmic characteristic life value, that is, the value in dB unit format is converted to a real value.

[0146] In summary, the model training method under high and low electrical stress conditions can be to first divide the high and low electrical stress conditions and then train the model separately, or the two methods can be used to perform global model training. After that, the areas of high and low electrical stress conditions are distinguished according to the results of model training. After simulation experiments, the life of power devices simulated by the two methods are basically the same, with small differences, and both can accurately fit the life of power devices. Figure 7The life of the power device evaluated by the model constructed in the embodiment of the present application can be seen to be well fitted with the actual life value under both high electrical stress conditions and low electrical stress conditions.

[0147] In the present application, a deep neural network model is used to train the high and low stress conditions respectively. Considering that the traditional prediction and evaluation model does not distinguish between high and low, the fitting effect is better under low electrical stress conditions, while the fitting effect is not good under high electrical stress conditions, and the fitting error is large. Therefore, by determining the high and low dividing points, the model is divided into two parts, and fitting calculations are performed separately for different electrical stress conditions, thereby improving the accuracy of the fitting.

[0148] The model constructed using the model building method of the present application scheme is used to perform failure analysis and life prediction on the new failure mechanisms and life characteristics exhibited by the newly developed GaN power devices. It is possible to identify the failure model of the new power device in advance and determine the failure mechanism of the new structure, thereby overcoming the rigid requirements of the traditional exponential function model and power function model for the characteristic life, and providing a theoretical basis for device reliability design.

[0149] Figure 1 FIG. 1 is a flow chart of a method for establishing a life assessment model in one embodiment. It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but the steps are not necessarily executed in the order indicated by the arrows; unless otherwise specified in this document, there is no strict order restriction for the execution of the steps, and the steps may be executed in other orders; and Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0150] Based on the above method, an embodiment of the present application also discloses a device for establishing a life assessment model.

[0151] Reference Figure 8 , the device includes the following modules:

[0152] A failure time calculation module 801 is used to determine the electrical stress condition of the power device according to the gate leakage current of the power device, and measure the failure time of the power device under the electrical stress condition;

[0153] The distribution function calculation module 802 is used to calculate the Weiss cumulative distribution function of the power device under the failure time;

[0154] The characteristic life calculation module 803 is used to calculate the characteristic life value of the power device under electrical stress conditions according to the Weiss cumulative distribution function and the preset global shape factor fitting;

[0155] The model building training module 804 is used to build a deep neural network model, and use the characteristic life value obtained by fitting calculation to train the deep neural network model to generate an accelerated life assessment model, which is a partitioned deep neural network model.

[0156] In one embodiment, the failure time calculation module 801 is specifically used to measure the steep increase point of the gate leakage current of the power device, and determine the steep increase stress range of the gate leakage current according to the steep increase point; determine several different electrical stress conditions according to the steep increase stress range; and use a dielectric breakdown full life accelerated experiment to measure the failure time of the power device under several electrical stress conditions.

[0157] In one embodiment, the distribution function calculation module 802 calculates the Weiss cumulative distribution function in the following manner:

[0158]

[0159] Where t is the failure time, β is the shape factor, η is the expansion factor, γ is the delay, and F(t) is the Weiss cumulative distribution function.

[0160] In one embodiment, the characteristic life calculation module 803 is specifically used to fit the Wei's distribution of failure time according to the Wei's cumulative distribution function; calculate the intercept value of the failure time under electrical stress conditions according to the Wei's distribution of failure time; calculate the characteristic life value of the power device under electrical stress conditions according to the preset global shape factor and the intercept value; the calculation method of the characteristic life value includes:

[0161]

[0162] Where intercept_j represents the calculated intercept value, β′ represents the global shape factor, and η_j represents the characteristic life value under the jth set of stress conditions.

[0163] In one embodiment, the model construction training module 804 is specifically used to construct the input layer, hidden layer and output layer of the deep neural network model, and set the activation function, weight and bias in the hidden layer; determine the demarcation point of the electrical stress condition according to the steep increase point of the gate leakage current, and generate a low electrical stress condition and a high electrical stress condition; determine the range of the electrical stress condition corresponding to the characteristic life value, and convert the format of the characteristic life value according to different electrical stress conditions; if the electrical stress condition corresponding to the characteristic life value is a low electrical stress condition, convert the characteristic life value into a real number format, and generate a real number characteristic life value; if the electrical stress condition corresponding to the characteristic life value is a high electrical stress condition, convert the characteristic life value into a dB unit format, and generate a logarithmic characteristic life value. life value; normalize the real characteristic life value and the logarithmic characteristic life value, and generate a real normalized characteristic life value and a logarithmic normalized characteristic life value; determine the range of the electrical stress condition corresponding to the characteristic life value, and train the deep neural network model according to different electrical stress conditions; if the electrical stress condition corresponding to the characteristic life value is a low electrical stress condition, the real normalized characteristic life value is used to train the deep neural network model, and a real neural network model is generated; if the electrical stress condition corresponding to the characteristic life value is a high electrical stress condition, the logarithmic normalized characteristic life value is used to train the deep neural network model, and a logarithmic neural network model is generated; the real neural network model and the logarithmic neural network model are set as accelerated life assessment models.

[0164] In one embodiment, the model construction training module 804 is specifically used to construct the input layer, hidden layer and output layer of the deep neural network model, and set the activation function, weight and bias in the hidden layer; convert the characteristic life value into real number format and dB unit format respectively, and generate the real characteristic life value and logarithmic characteristic life value after the format conversion; normalize the real characteristic life value and logarithmic characteristic life value after the format conversion, and generate the real normalized characteristic life value and logarithmic normalized characteristic life value; use the real normalized characteristic life value to train the deep neural network model and generate the real neural network model; use the logarithmic normalized characteristic life value to train the deep neural network model and generate the logarithmic neural network model; according to the training results of the real neural network model and the logarithmic neural network model, divide the electrical stress conditions into low electrical stress conditions and high electrical stress conditions; use the real neural network model correspondingly under the low electrical stress condition, and use the logarithmic neural network model correspondingly under the high electrical stress condition, and generate the final accelerated life assessment model.

[0165] In one embodiment, the model building training module 804 is also used to perform a denormalization process on the results output by the accelerated life assessment model.

[0166] The life assessment model establishment device provided in the embodiment of the present application can be applied to the life assessment model establishment method provided in the above embodiment. For relevant details, refer to the above method embodiment. The implementation principle and technical effect are similar and will not be repeated here.

[0167] It should be noted that: the life assessment model establishment device provided in the embodiment of the present application only uses the division of the above-mentioned functional modules / functional units as an example when establishing a life assessment model. In actual applications, the above-mentioned functions can be assigned to different functional modules / functional units as needed, that is, the internal structure of the life assessment model establishment device is divided into different functional modules / functional units to complete all or part of the functions described above. In addition, the implementation method of the life assessment model establishment method provided in the above method embodiment and the implementation method of the life assessment model establishment device provided in this embodiment belong to the same concept. The specific implementation process of the life assessment model establishment device provided in this embodiment is detailed in the above method embodiment, and will not be repeated here.

[0168] The embodiment of the present application also discloses a computer device.

[0169] Specifically, if Fig. 9 As shown, the computer device can be a computer device such as a desktop computer, a laptop computer, a PDA, and a cloud server. The computer device may include, but is not limited to, a processor and a memory. The processor and the memory may be connected via a bus or otherwise. The processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, graphics processors (GPU), embedded neural network processors (NPU) or other dedicated deep learning coprocessors, discrete gates or transistor logic devices, discrete hardware components and other chips, or a combination of the above-mentioned various chips.

[0170] As a non-transient computer-readable storage medium, the memory can be used to store non-transient software programs, non-transient computer executable programs and modules, such as program instructions / modules corresponding to the methods in the above-mentioned embodiments of the present application. The processor executes various functional applications and data processing of the processor by running the non-transient software programs, instructions and modules stored in the memory, that is, the method in the above-mentioned method implementation is realized. The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0171] The embodiment of the present application also discloses a computer-readable storage medium.

[0172] Specifically, a computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method in the above-mentioned method implementation is implemented. Those skilled in the art will understand that the implementation of all or part of the process in the above-mentioned implementation method of the present application can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and the program, when executed, may include the process of the implementation of the above-mentioned methods. Among them, the storage medium may be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk (Hard Disk Drive, abbreviated as: HDD) or a solid-state drive (SSD), etc.; the storage medium may also include a combination of the above-mentioned types of memory.

[0173] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A method for establishing a life assessment model, characterized in that: The method is applied to a life assessment model establishment system, the life assessment model establishment system includes a power device, and the method includes: determining an electrical stress condition of the power device according to a gate leakage current of the power device, and measuring a failure time of the power device under the electrical stress condition; Calculate the Wei's cumulative distribution function of the power device at the failure time; Calculate the characteristic life value of the power device under the electrical stress condition according to the Weiss cumulative distribution function and the preset global shape factor fitting; A deep neural network model is constructed, and the characteristic life value obtained by fitting calculation is used to train the deep neural network model to generate an accelerated life assessment model, where the accelerated life assessment model is a partitioned deep neural network model.

2. The method according to claim 1, characterized in that: Determining the electrical stress condition of the power device according to the gate leakage current of the power device, and measuring the failure time of the power device under the electrical stress condition comprises: Measuring a steep increase point of the gate leakage current of the power device, and determining a steep increase stress range of the gate leakage current according to the steep increase point; Determining a number of different electrical stress conditions according to the steep stress range; The dielectric breakdown full life acceleration experiment is used to measure the failure time of the power device under several electrical stress conditions.

3. The method according to claim 1, characterized in that: The calculation method of the Wei's cumulative distribution function includes: Wherein, t represents the failure time, β represents the shape factor, η represents the expansion factor, γ represents the delay, and F(t) represents the Weiss cumulative distribution function.

4. The method according to claim 1, characterized in that: The calculating the characteristic life value of the power device under the electrical stress condition according to the Weiss cumulative distribution function and the preset global shape factor fitting includes: Fitting the Wei's distribution of the failure time according to the Wei's cumulative distribution function; Calculating the intercept value of the failure time under the electrical stress condition according to the Wei's distribution of the failure time; Calculating a characteristic life value of the power device under the electrical stress condition according to the preset global shape factor and the intercept value; The calculation method of the characteristic life value includes: Wherein, intercept_j represents the calculated intercept value, β′ represents the global shape factor, and η_j represents the characteristic life value under the jth set of stress conditions.

5. The method according to claim 2, characterized in that: The step of constructing a deep neural network model and training the deep neural network model using the characteristic life value obtained by fitting calculation to generate an accelerated life assessment model comprises: Constructing the input layer, hidden layer and output layer of the deep neural network model, and setting activation functions, weights and biases in the hidden layer; Determining a demarcation point of the electrical stress condition according to a steep increase point of the gate leakage current, and generating a low electrical stress condition and a high electrical stress condition; Determine the range of the electrical stress condition corresponding to the characteristic life value, and convert the format of the characteristic life value according to different electrical stress conditions; If the electrical stress condition corresponding to the characteristic life value is a low electrical stress condition, converting the characteristic life value into a real number format and generating a real number characteristic life value; If the electrical stress condition corresponding to the characteristic life value is a high electrical stress condition, converting the characteristic life value into a dB unit format and generating a logarithmic characteristic life value; Normalizing the real characteristic life value and the logarithmic characteristic life value to generate a real normalized characteristic life value and a logarithmic normalized characteristic life value; Determining a range of electrical stress conditions corresponding to the characteristic life value, and training the deep neural network model according to different electrical stress conditions; If the electrical stress condition corresponding to the characteristic life value is a low electrical stress condition, the deep neural network model is trained using the real normalized characteristic life value to generate a real neural network model; If the electrical stress condition corresponding to the characteristic life value is a high electrical stress condition, the deep neural network model is trained using the logarithmic normalized characteristic life value to generate a logarithmic neural network model; The real number neural network model and the logarithmic neural network model are set as the accelerated lifetime assessment model.

6. The method according to claim 1, characterized in that: The step of constructing a deep neural network model and training the deep neural network model using the characteristic life value obtained by fitting calculation to generate an accelerated life assessment model comprises: Constructing the input layer, hidden layer and output layer of the deep neural network model, and setting activation functions, weights and biases in the hidden layer; Converting the characteristic life value into a real number format and a dB unit format respectively, and generating a real number characteristic life value and a logarithmic characteristic life value after the format conversion; Normalizing the real number characteristic life value and the logarithmic characteristic life value after format conversion, and generating a real number normalized characteristic life value and a logarithmic normalized characteristic life value; Using the real normalized characteristic life value to train the deep neural network model and generate a real neural network model; Using the logarithmic normalized characteristic life value to train the deep neural network model and generate a logarithmic neural network model; According to the training results of the real neural network model and the logarithmic neural network model, the electrical stress condition is divided into a low electrical stress condition and a high electrical stress condition; Under the low electrical stress condition, the real number neural network model is used accordingly, and under the high electrical stress condition, the logarithmic neural network model is used accordingly, and the final accelerated life assessment model is generated.

7. The method according to claim 5 or 6, characterized in that: After the accelerated life assessment model is generated, the method further includes: performing a denormalization process on a result output by the accelerated life assessment model.

8. A device for establishing a life assessment model, characterized in that: The device is applied to a life assessment model establishment system, the life assessment model establishment system includes a power device, and the device includes: A failure time calculation module (801) is used to determine the electrical stress condition of the power device according to the gate leakage current of the power device, and measure the failure time of the power device under the electrical stress condition; A distribution function calculation module (802), used to calculate the Weiss cumulative distribution function of the power device at the failure time; A characteristic life calculation module (803), used for calculating the characteristic life value of the power device under the electrical stress condition according to the Weiss cumulative distribution function and a preset global shape factor fitting; The model building training module (804) is used to build a deep neural network model and train the deep neural network model using the characteristic life value obtained by fitting calculation to generate an accelerated life assessment model, wherein the accelerated life assessment model is a partitioned deep neural network model.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 7.

Images